Method for constructing large industrial chain model

Abstract

The invention provides a method for constructing an industrial chain large model, and relates to the field of big data model construction calculation. The method comprises the following steps: S1, data acquisition and arrangement: collecting data from each link of an industrial chain; s2, data annotation: annotating the data through an annotation software flywheel annotation system; s3, model selection and framework design: selecting a composite system collaborative model as a model scheme according to the property of the problem; s4, carrying out model training and optimization, and carrying out pre-training on the model by using large-scale unsupervised data; s5, deploying and applying the model, deploying the industrial chain large model into the production environment, reducing the probability that the process is high in cost and long in time consumption due to the fact that the original standardized design is possibly not applicable any more, needs to collect and arrange data again and reconstructs the logical relationship among the data, improving the collaboration degree among the subsystems, and improving the efficiency of the system. Resource configuration is optimized, and redundancy and waste are reduced.

Description

Technical Field

[0001] The present invention relates to the field of big data model construction and calculation, and in particular to a method for constructing a large industrial chain model. Background Art

[0002] The industrial chain refers to the entire process from the supply of raw materials to the production, sales and services of final products in a specific industry. The industrial chain can be divided into three parts: upstream, midstream and downstream. The upstream is mainly the suppliers of raw materials and parts, the midstream is the product manufacturing and assembly links, and the downstream is the product sales and after-sales service links.

[0003] When constructing the existing industrial chain model, for the selected target industrial chain, the vertical industrial chain and related industrial chain of the target industrial chain are obtained based on the comprehensive analysis of the production and application of big data. Then, based on the level of data production, the different levels and applications of big data, a big data component application framework model is established, and a big data structure morphology classification model is established. Finally, according to the characteristics of the industrial ecosystem, the target industrial chain is divided into links. At the same time, the business process comprehensive data structure and the industrial subject comprehensive data structure of the target industrial chain are standardized. Combined with the manual interaction data, the big data component application framework model and the big data structure morphology classification model in the modeling process, the logical relationship between the data is constructed to complete the establishment of the industrial chain ecological big data model. However, when the existing method of constructing the industrial chain model is used, once the big data component application framework model and the big data structure morphology classification model are determined, it is relatively difficult to adjust in the face of dynamic changes in the industrial chain. When the target industrial chain is impacted by new technologies or market demand changes suddenly, the original standardized design may no longer be applicable, and it is necessary to collect and organize data again and reconstruct the logical relationship between data. This process is costly and time-consuming.

[0004] Therefore, it is necessary to provide a new method for constructing a large industrial chain model to solve the above technical problems. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method for constructing a large industrial chain model.

[0006] A method for constructing a large industrial chain model provided by the present invention comprises the following steps: S1. Data collection and collation: collect data from all links of the industrial chain, pre-process the collected data, and classify the data into unsupervised data and supervised data according to specific tasks and model requirements; S2, data annotation, annotating the data through the annotation software Flywheel Annotation System; S3. Model selection and framework design. According to the nature of the problem, the composite system collaborative model is selected as the model solution. The composite system collaborative model includes the industrial chain and innovation chain Two subsystems, industrial chain and innovation chain Build a collaborative communication mechanism between them, and combine the business logic and data relationship of the industrial chain to build a large model of the industrial chain; S4, model training and optimization, using the unsupervised data collected from various industry chains to pre-train the model for language model tasks and image-related tasks, and based on the pre-training, fine-tune the model with supervised data related to the industry chain; S5. Model deployment and application: deploy the large industrial chain model into the production environment, and continuously optimize and update the model based on real-time collected data.

[0007] Preferably, the workflow of the composite system synergy model includes the following steps: Step 1: System analysis and subsystem division: by determining the scope of the industrial chain complex system under study and dividing the complex system into multiple interrelated subsystems according to the structural and functional characteristics of the industrial chain; Step 2: Variable selection and indicator system construction: For each subsystem, select key variables that can describe its state and behavior. At the same time, based on the selected state variables, build an indicator system to measure the degree of coordination between subsystems; Step 3: Data collection and preprocessing: Collect data from enterprises, institutions and related platforms involved in each subsystem of the industrial chain, clean the collected data, and remove noise data, duplicate data and erroneous data; Step 4: Construct and calculate the synergy model. According to the characteristics and data of the industrial chain complex system, select a suitable synergy calculation model and determine the parameters of the selected synergy model. Step 5: Synergy analysis and structural interpretation: Based on the calculated synergy value, the synergy level of the industrial chain complex system is evaluated, and the subsystems or relationships between subsystems with low synergy are analyzed.

[0008] Preferably, the various links of the industrial chain described in step S1 include the upstream link - raw material supply and primary processing, the midstream link - product manufacturing and assembly, and the downstream link - product sales and after-sales service.

[0009] Preferably, the marking software flywheel marking system in step S2 includes the following steps when marking: Step 1: Data import and preparation stage: The data related to the industrial chain needs to be converted into a format that can be recognized by the Flywheel Annotation System, and then the converted data is imported into the Flywheel Annotation System. After the data is imported, the data is classified and grouped according to the links of the industrial chain or the purpose of the data; Step 2: Select and set annotation tools. The Flywheel annotation system provides a variety of annotation tools. You can select annotation tools based on the data type and set the tool parameters according to the specific annotation task. Step 3: During the labeling process, the labeler labels the product according to the preset labeling tools and rules through the user interface of the flywheel labeling system; Step 4: Annotation structure management and export. After the annotation is completed, the annotation results need to be reviewed and the reviewed annotation data needs to be exported from the Flywheel annotation system.

[0010] Preferably, the unsupervised data in step S4 includes text data, image data, time series data and relational data, and the supervised data includes product quality related data, market and sales data, supply chain data and enterprise operation and management data.

[0011] Preferably, the applications and tools in step S5 include data processing and analysis tools, model development and training frameworks, model deployment platforms, and application development platforms and interface tools.

[0012] Compared with the related art, the method for constructing a large industrial chain model provided by the present invention has the following beneficial effects: The present invention constructs a large model of the industrial chain through a composite system synergy model. By determining the scope of the industrial chain composite system under study and dividing the composite system into multiple interrelated subsystems according to the structure and functional characteristics of the industrial chain, for each subsystem, key variables that can describe its state and behavior are selected. At the same time, based on the selected state variables, an indicator system for measuring the degree of synergy between subsystems is constructed, data is collected from enterprises, institutions and related platforms involved in each subsystem of the industrial chain, the collected data is cleaned, noise data, duplicate data and erroneous data are removed, and a suitable synergy calculation model is selected according to the characteristics and data conditions of the industrial chain composite system. At the same time, the parameters of the selected synergy model are determined, and the synergy level of the industrial chain composite system is evaluated according to the calculated synergy value. At the same time, subsystems with low synergy or the relationship between subsystems are analyzed, which reduces the possibility that the original standardized design may no longer be applicable, and the need to re-collect and organize data and re-construct the logical relationship between data, a process with high cost and long time consumption, and improves the degree of synergy between subsystems, optimizes resource allocation, and reduces redundancy and waste. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A flowchart of a method for constructing a large industrial chain model provided by the present invention; Figure 2 A flowchart of the composite system coordination model provided by the present invention; Figure 3 This is a flowchart of the flywheel annotation system, an annotation software provided by the present invention. DETAILED DESCRIPTION

[0014] The present invention will be further described below in conjunction with the accompanying drawings and implementation modes.

[0015] Please refer to Figure 1 , Figure 2 as well as Figure 3 ,in, Figure 1 A flowchart of a method for constructing a large industrial chain model provided by the present invention; Figure 2 A flowchart of the composite system coordination model provided by the present invention; Figure 3 This is a flowchart of the flywheel annotation system, an annotation software provided by the present invention.

[0016] In the specific implementation process, Figures 1 to 3 As shown, the following steps are included; S1. Data collection and collation: collect data from all links of the industrial chain, pre-process the collected data, and classify the data into unsupervised data and supervised data according to specific tasks and model requirements; It should be noted that the various links of the industrial chain in step S1 include the upstream link—raw material supply and primary processing, the midstream link—product manufacturing and assembly, and the downstream link—product sales and after-sales service; Among them, the upstream link - raw material supply and primary processing includes raw material collection data and primary processing data, principle collection data, including the geographical distribution information of minerals, agricultural raw materials, and forest resources, and records the quantity, speed, and completion of the mining plan of raw materials, primary processing data, process parameters of raw materials during preliminary processing, such as temperature, pressure, time, and amount of chemical agents, and data obtained by testing the quality of products after primary processing, including purity, particle size, and strength indicators; Midstream links - product manufacturing includes parts manufacturing data and product assembly data. Parts manufacturing data records the design information of parts, including size, shape, material requirements, and the production process arrangement, equipment usage, and worker operation records during parts manufacturing. Product assembly data, the process flow, assembly sequence, and connection method during product assembly. At the same time, the data obtained from quality inspection of semi-finished products and final products during the assembly process, such as functional test results and appearance inspection records; Downstream links - product sales and after-sales service include market and sales data and after-sales service data. Market and sales data include market size, market growth rate, consumer demand preferences, and record product sales channel information, such as sales data of online e-commerce platforms, offline specialty stores, dealers, etc., after-sales service data, customer complaints and suggestions on product quality, performance, user experience, etc., as well as product maintenance records, maintenance parts usage, and maintenance cycle data; At the same time, when preprocessing the data, the data should be cleaned first. Some data may be missing in the collected data. For missing values, a variety of methods can be used to deal with them, such as deleting records with missing values ​​(applicable when the missing data accounts for a small proportion and there are many data records), filling with mean, median or mode (applicable to numerical data), or using more complex machine learning algorithms for prediction and filling. When dealing with outliers, statistical methods (such as 3D CNN) can be used to identify outliers. Principle, that is, data values ​​outside the range of mean ±3 times standard deviation are considered as outliers) are identified, and then corrected or deleted according to the actual situation to remove duplicate data. By comparing the key fields of the data (such as transaction time, stock code, transaction amount, etc.), duplicate data can be identified and deleted; Data conversion: Data standardization. In order to effectively process data from different sources and different magnitudes in a unified model, the data needs to be standardized. Common standardization methods include Z-score standardization (converting data to a distribution with a mean of 0 and a standard deviation of 1) and Min-Max standardization (mapping data to the interval [0, 1]). Data coding: some non-numeric data, such as product categories and customer level classification data, need to be coded; Data aggregation and splitting: Aggregate and split data according to the needs of the model. In the supply chain industry chain, in order to analyze the monthly inventory turnover rate, daily inventory data and sales data can be aggregated to calculate the relevant indicators for each month. On the contrary, sometimes it is also necessary to split some comprehensive data into finer granularity; S2, data annotation, annotating the data through the annotation software Flywheel Annotation System; It should be noted that the marking software flywheel marking system in step S2 includes the following steps when marking: Step 1: Data import and preparation stage: The data related to the industrial chain needs to be converted into a format that can be recognized by the Flywheel Annotation System, and then the converted data is imported into the Flywheel Annotation System. After the data is imported, the data is classified and grouped according to the links of the industrial chain or the purpose of the data; Step 2: Select and set annotation tools. The Flywheel annotation system provides a variety of annotation tools. You can select annotation tools based on the data type and set the tool parameters according to the specific annotation task. Step 3: During the labeling process, the labeler labels the product according to the preset labeling tools and rules through the user interface of the flywheel labeling system; Step 4: Annotation structure management and export. After the annotation is completed, the annotation results need to be reviewed, and the reviewed annotation data should be exported from the Flywheel annotation system. Among them, the Flywheel Annotation System is a data annotation tool or platform that aims to provide efficient and accurate annotation services for various data types to meet the needs of machine learning, artificial intelligence and other fields for high-quality annotated data. Its core lies in converting raw data into annotated data that machines can understand and learn through a series of technical means and workflows, thereby promoting the training and optimization of related models. S3. Model selection and framework design. According to the nature of the problem, the composite system collaborative model is selected as the model solution. The composite system collaborative model includes the industrial chain and innovation chain Two subsystems, industrial chain and innovation chain Build a collaborative communication mechanism between them, and combine the business logic and data relationship of the industrial chain to build a large model of the industrial chain; It should be noted that the workflow of the composite system synergy model includes the following steps: Step 1: System analysis and subsystem division: by determining the scope of the industrial chain complex system under study and dividing the complex system into multiple interrelated subsystems according to the structural and functional characteristics of the industrial chain; Step 2: Variable selection and indicator system construction: For each subsystem, select key variables that can describe its state and behavior. At the same time, based on the selected state variables, build an indicator system to measure the degree of coordination between subsystems; Step 3: Data collection and preprocessing: Collect data from enterprises, institutions and related platforms involved in each subsystem of the industrial chain, clean the collected data, and remove noise data, duplicate data and erroneous data; Step 4: Construct and calculate the synergy model. According to the characteristics and data of the industrial chain complex system, select a suitable synergy calculation model and determine the parameters of the selected synergy model. Step 5: Synergy analysis and structural interpretation: Based on the calculated synergy value, the synergy level of the industrial chain complex system is evaluated, and the subsystems or the relationship between subsystems with low synergy are analyzed; The subsystem is composed of several order parameters Composition, of which , , and The parameters are The lower and upper limits of At the same time, the calculation process of the order of the composite system includes the calculation of the order parameter order, the calculation of the subsystem order, and the calculation of the composite system order. The calculation method includes some calculation formulas: In the formula, The value of the order parameter represented by is proportional to the order degree of its order parameter, that is, the larger the value, the higher the order degree of the order parameter. The value of the order parameter represented by is inversely proportional to the order degree of its order parameter, that is, the smaller the value, the higher the order degree of the order parameter; Each subsystem The order degree measurement of each subsystem is realized by integrating the order degree of the order parameters of each subsystem, and the integration is performed using the linear weighted method. The formula for calculating the order degree is: Since the calculation formula of subsystem order degree can be known The order of the "two-chain" composite system is calculated based on the change in the order of each order parameter between two moments. The order of each subsystem is , at the moment of the evolution of the complex system The order of each subsystem is , then the test formula for the synergy of the “two-chain” composite system is: in, , j=1,2; The collaborative communication mechanism is to hold regular exchange meetings and workshops for relevant departments of the industrial chain and innovation chain, so that personnel from production, sales, R&D and other departments can participate together to strengthen communication and collaboration, and build an information sharing platform for the industrial chain and innovation chain to achieve real-time data sharing; S4, model training and optimization, using the unsupervised data collected from various industry chains to pre-train the model for language model tasks and image-related tasks, and based on the pre-training, fine-tune the model with supervised data related to the industry chain; It should be noted that the unsupervised data in step S4 includes text data, image data, time series data and relational data, and the supervised data includes product quality related data, market and sales data, supply chain data and enterprise operation and management data; At the same time, when training the model, data division is first performed. Before starting training, the collected and organized data is usually divided into training set, validation set and test set. The training set is used to learn and adjust the model parameters. The validation set is used to evaluate the performance of the model during the training process to prevent overfitting and to adjust the hyperparameters of the model. The test set is used to evaluate the final performance of the model after the model training is completed. Then the training algorithm is selected. For most machine learning and deep learning models, gradient descent is a commonly used optimization algorithm. Stochastic gradient descent randomly selects a sample each time for gradient calculation and parameter update. It is fast but may cause unstable convergence. Then set the training parameters. The hyperparameters include learning rate, number of iterations, number of neural network layers and number of neurons in each layer. Gradient descent is an optimization algorithm that finds the minimum value of a function based on the function gradient. In model training, we usually have a loss function that measures the difference between the model's predicted results and the actual results. The goal of gradient descent is to minimize the value of the loss function by continuously adjusting the model's parameters; A simple linear regression model For example, assume the loss function is: Where m is the number of samples, and is the true value and characteristic value of the i-th sample; The gradient descent algorithm calculates the loss function with respect to the parameters and The partial derivative of and , the vector composed of these partial derivatives is the gradient. The direction of the gradient indicates the direction in which the loss function grows fastest, and the opposite direction of the gradient is the direction in which the loss function decreases fastest; And, in each iteration, the parameters are updated according to the following formula: (j=0,1), where is the learning rate, which determines the step size of each parameter update; S5, model deployment and application, deploying the large industrial chain model into the production environment, and continuously optimizing and updating the model based on real-time collected data; It should be noted that the applications and tools in step S5 include data processing and analysis tools, model development and training frameworks, model deployment platforms, and application development platforms and interface tools; Model application: In the upstream of the industrial chain, the industrial chain model can help enterprises better manage suppliers, and in the R&D link of the upstream of the industrial chain, the model can be used to evaluate the technical capabilities and R&D potential of different suppliers; In the application of midstream industrial chain, in the product manufacturing link, the industrial chain big model can optimize the production process, and can also timely discover quality problems and make adjustments through real-time analysis of quality data in the production process. At the same time, for the inventory management of midstream enterprises, the model can predict the market demand and inventory turnover rate of products; In the downstream application of the industrial chain, the model can be used for market segmentation and target customer positioning. At the same time, the model can also predict the sales effects of different sales channels, help companies optimize the sales channel layout, and use the large industrial chain model to better handle customer feedback.

[0017] The circuits and controls involved in the present invention are all prior art and will not be described in detail here.

[0018] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for constructing a large industrial chain model, characterized in that: The steps include: S1. Data collection and collation: collect data from all links of the industrial chain, pre-process the collected data, and classify the data into unsupervised data and supervised data according to specific tasks and model requirements; S2, data annotation, annotating the data through the annotation software Flywheel Annotation System; S3. Model selection and framework design. According to the nature of the problem, the composite system collaborative model is selected as the model solution. The composite system collaborative model includes the industrial chain and innovation chain Two subsystems, industrial chain and innovation chain Build a collaborative communication mechanism between them, and combine the business logic and data relationship of the industrial chain to build a large model of the industrial chain; S4, model training and optimization, using the unsupervised data collected from various industry chains to pre-train the model for language model tasks and image-related tasks, and based on the pre-training, fine-tune the model with supervised data related to the industry chain; S5. Model deployment and application: deploy the large industrial chain model into the production environment, and continuously optimize and update the model based on real-time collected data.

2. The method for constructing a large industrial chain model according to claim 1, characterized in that: The workflow of the composite system synergy model includes the following steps: Step 1: System analysis and subsystem division: by determining the scope of the industrial chain complex system under study and dividing the complex system into multiple interrelated subsystems according to the structural and functional characteristics of the industrial chain; Step 2: Variable selection and indicator system construction: For each subsystem, select key variables that can describe its state and behavior. At the same time, based on the selected state variables, build an indicator system to measure the degree of coordination between subsystems; Step 3: Data collection and preprocessing: Collect data from enterprises, institutions and related platforms involved in each subsystem of the industrial chain, clean the collected data, and remove noise data, duplicate data and erroneous data; Step 4: Construct and calculate the synergy model. According to the characteristics and data of the industrial chain complex system, select a suitable synergy calculation model and determine the parameters of the selected synergy model. Step 5: Synergy analysis and structural interpretation: Based on the calculated synergy value, the synergy level of the industrial chain complex system is evaluated, and the subsystems or relationships between subsystems with low synergy are analyzed.

3. The method for constructing a large industrial chain model according to claim 2, characterized in that: The various links of the industrial chain described in step S1 include the upstream link - raw material supply and primary processing, the midstream link - product manufacturing and assembly, and the downstream link - product sales and after-sales service.

4. The method for constructing a large industrial chain model according to claim 3, characterized in that: The marking software flywheel marking system in step S2 includes the following steps when marking: Step 1: Data import and preparation stage: The data related to the industrial chain needs to be converted into a format that can be recognized by the Flywheel Annotation System, and then the converted data is imported into the Flywheel Annotation System. After the data is imported, the data is classified and grouped according to the links of the industrial chain or the purpose of the data; Step 2: Select and set annotation tools. The Flywheel annotation system provides a variety of annotation tools. You can select annotation tools based on the data type and set the tool parameters according to the specific annotation task. Step 3: During the labeling process, the labeler labels the product according to the preset labeling tools and rules through the user interface of the flywheel labeling system; Step 4: Annotation structure management and export. After the annotation is completed, the annotation results need to be reviewed and the reviewed annotation data needs to be exported from the Flywheel annotation system.

5. The method for constructing a large industrial chain model according to claim 4, characterized in that: The unsupervised data in step S4 includes text data, image data, time series data and relational data, and the supervised data includes product quality related data, market and sales data, supply chain data and enterprise operation and management data.

6. The method for constructing a large industrial chain model according to claim 5, characterized in that: The applications and tools described in step S5 include data processing and analysis tools, model development and training frameworks, model deployment platforms, and application development platforms and interface tools.

Images